Software bug reports often miss critical information, delaying their resolution. The last decade has seen a growing trend of combining textual and multimedia information from software artifacts (e.g., bug reports, and programming questions) to support various software engineering tasks (e.g., duplicate bug report detection, and bug reproduction). However, none of the studies performs a fine-grained analysis of the multimedia information attached to bug reports. Hence, it is not clear what their attached images or videos contain or whether they could help identify software bugs or errors automatically. In this paper, we conduct a preliminary study that investigates the presence or prevalence of key elements in 1,469 images attached to 1,000 bug reports and demonstrate their potential to support IR-based bug localization. We have several interesting findings. First, our analysis suggests that the attached images to visual bug reports contain a mix of UI components, programming components, and regular text. Second, our analysis using an LLM (e.g., GPT4o) suggests that it can extract the key elements from the attached images effectively, posing a suitable alternative to human annotators. Finally, our experiments suggest that the multimedia information extracted from the attached images can enhance the performance of a traditional technique for bug localization by improving 34.06 % of its search queries.

Over the past few decades, underwater image enhancement has attracted an increasing amount of research effort due to its significance in underwater robotics and ocean engineering. Research has evolved from implementing physics-based solutions to using very deep CNNs and GANs. However, these state-of-art algorithms are computationally expensive and memory intensive. This hinders their deployment on portable devices for underwater exploration tasks. These models are trained on either synthetic or limited real-world datasets making them less practical in real-world scenarios. In this paper, we propose a shallow neural network architecture, Shallow-UWnet which maintains performance and has fewer parameters than the state-of-art models. We also demonstrated the generalization of our model by benchmarking its performance on a combination of synthetic and real-world datasets.